Braided river reaches and alluvial systems are abundant in many areas, they are characterised by their multi-threaded planform and are agents of substantial sediment transport, erosion and deposition. The high rates of sediment transport, erosion and deposition, and the frequent shifting of river channel positions in braided rivers pose many interesting problems to a whole range of disciplines. Despite this importance they have been relatively neglected in academic study when compared with the wealth of material on meandering rivers (Bristow and Best, 1993). The majority of studies to-date have been qualitative in nature, with Howard et al. (1970) and Murray and Paola (1994) being notable exceptions. Only recently have more quantitative studies been carried out. This relative disregard of braided rivers has resulted in the development of our understanding of braided river behaviour being impeded.

The neglect of braided river study is probably due in part to the difficulty of undertaking field work and characterising complex features in this rapidly changing environment. Although advances have been made in the qualitative understanding of flow and sediment processes in braided systems, Bristow and Best (1993) identified several key issues remain to be addressed, they include; (a) the mechanisms of braid bar genesis and evolution, (b) flow and sediment dynamics at bifurcations and confluences, (c) the influence of flow stage on planform development, (d) the implications of a channel hierarchy system found over a range of channel scales and (e) the influence of secondary currents on the morphological development of braid bars.


A natural river will adjust its channel pattern in space along its channel network and in time at a given point to accommodate imposed flow and sediment regimes. The forms of a natural channels when viewed in plan fall within a continuum of channel patterns that is traditionally sub-divided into straight, meandering, braided and anastomosed. The term 'braided' has been given several definitions in academic literature over the past 40 years (Bridge, 1993). Leopold and Wolman (1957) described the braided river as 'one which flows in two or more anastomosing channels around alluvial islands' while Lane (1957) reported that 'a braided steam is characterised by having a number of alluvial channels with bars or islands between meeting and dividing again, and presenting from the air the intertwining effect of a braid'. Brice (1964) recognised the importance of defining the difference between mid-channel bars or islands within braided rivers and portions of the floodplain excised by channel diversions and avulsions. Schumm (1977) made the distinction between braided rivers that at low stages have islands of sediment or islands of semi-permanent vegetation, and multiple-thread rivers or anastomosing channels that have branches with individual channel patterns.

Bridge (1993) concluded that these conflicting definitions of the braided pattern raise the issues concerning; (a) the difference between mid-channel bars and islands, (b) the precise nature of the interaction between flow stage and bars or islands and (c) the differences between the mechanisms of channel divergence that lead to river patterns termed as 'braided' and those defined as 'anastomosing'.

1.2.1 Classification Systems for River Channel Pattern

Bridge (1993) found that Leopold and Wolman's (1957) classification system for river pattern was inappropriate. Leopold and Wolman (1957) classified river pattern to be either straight, meandering or braided, each class being separated by threshold values of river discharge and channel slope (figure 1.1). They regarded a single channel division around a bar or island as a braid, and showed that hydraulic factors such as slope adjusted to the presence of braids. Meandering channels were defined as having a sinuosity greater than or equal to a value of 1.5, whereas channel multiplicity defined the braided pattern. Bridge (1993) stated that 'this is unsatisfactory because the classes are not mutually exclusive and different parameters are used to define the different patterns' (page 19). As a result, a channel may contain characteristics of both channel patterns, in that it may by a system with a sinuosity greater than 1.5 which also contains alluvial islands or bars (Rust, 1978). Also, as reported by Knighton and Nanson (1993), Leopold and Wolman's scheme does not explicitly recognise the anastomosed channel pattern.

Figure 1. Function distinguishing between meanders and braided channels on the basis of channel slope and discharge (Leopold and Wolman, 1957)

Rust (1978) proposed a system of alluvial channel classification based on two parameters: the braiding index and sinuosity. The braiding index was defined by the number of bars per mean meander wavelength, with single and multi-channel systems having a braiding index of less than and greater than one, respectively. The channels are then divided further into two categories of high and low sinuosity, with the distinction made at a sinuosity of 1.5. This classification lead to four channel types: (i) single-channel high-sinuosity (meandering); (ii) single-channel low-sinuosity (straight); (iii) multi-channel high-sinuosity (anastomosing); (iv) multi-channel low-sinuosity (braided).

To identify more clearly the anastomosing channel pattern, Knighton and Nanson (1993) suggested a classification system for channel pattern that did not include sinuosity as a parameter. Their classification system is based on a continuum concept using three variables; flow strength, bank erodibility and relative sediment supply. Straight reaches are represented by low values of flow strength, bank erodibility and sediment supply. Meandering reaches are represented by medium values of flow strength and low to medium values of bank erodibility and sediment supply. Braided reaches are defined as those in which there is a high flow strength, high bank erodibility and medium to high sediment supply, as opposed to anastomosed reaches which are defined as having low flow strength, low bank erodibility and medium to high sediment supply. Knighton and Nanson (1993) stated that sinuosity is probably not a sufficiently discriminatory characteristic for classifying any channel pattern, but particularly anastomosed channels, because it can vary widely not only between anastomosing rivers, but also between different channels within a given anastomosing reach.

Robertson-Rintoul and Richards (1993) stated that classification systems similar to that developed by Leopold and Wolman (1957) that separated meandering and braided channel patterns based on threshold values of discharge (Q) and slope (S) are flawed because both the coefficient and constant in the power relationship between discharge and slope vary with bed material size (Henderson, 1961), bank sediment (Ferguson, 1984; 1987) and the discharge criterion employed (Antropovskiy, 1972). They also commented that assigning a nominal planform class to a river system, such as 'meandering' or 'braided', implied that there is greater uniformity of morphology within a class than is the case in reality.

Robertson-Rintoul and Richards proposed a classification system based on an index of total sinuosity, building on earlier work by Richards (1982). Their approach involves the quantitative evaluation of parameters directly related to meandering and braided patterns and the description of how they vary with controlling variables such as stream power. They found that the following relationship existed for sand and gravel-bed rivers respectively:

SP = 1 + 2.64(Qsv)0.38 D84 -0.44 (equation 1.1)

SP = 1 + 5.52(Qsv)0.40 D84 -0.14 (equation 1.2)


SP = index of total sinuosity (cumulative length of all channels / reach length

Q = discharge (of return period 1.5-2.3 years)

sv = valley slope

D84 = grain size for which 84% of material is finer.

Bridge (1993), also recognised the shortfalls of previous attempts at a universal classification system for channel pattern and suggested that any descriptive classification system should first incorporate the nature of channel splitting around bars or islands and then the degree of sinuosity in channel(s).

1.2.2 The Distinction Between Bars and Islands

The problem of how best to differentiate between bars and islands within braided and anastomosed channel patterns has yet to be fully addressed. Brice (1964) defined bars as transient features that are submerged at bankfull stage and are unvegetated, whereas islands are stabilised features that are vegetated and are not submerged at bankfull stage. Bridge (1993) stated that the difficultly with the definition of bars and islands put forward by Brice (1964) is that the degree of vegetation that exists on a mid-channel feature is dependent on the length of time that feature remains emergent above the flow stage, the type of vegetation that is available for colonization, the nature of available sediment and the climate. Bridge (1993) further suggested that in order to overcome the subjective nature of terms such as 'transient', 'stable' and 'unstable', they should be replaced by quantitative measures of bar persistence, rates of deposition, migration and erosion.

1.2.3 The Influence of Flow Stage on Channel Pattern Classification

Another key issue in defining the braided channel pattern is the influence of flow stage. Generally, the form of bars is controlled by high in-bank flows, although some features such as cross-bar channels are controlled by falling stages (Bristow, 1987). Bridge (1993) recommends that the braided channel pattern should ideally be observed and characterised at formative high discharges when falling stage features are not present. Whilst accepting that it can sometimes be difficult to achieve in practice. Kellerhals et al. (1976) and Rust (1978) recommend the definition of channel patterns at mid-way or mean channel stages that have a high frequency of occurrence.

A further problem in defining river channel pattern examined by Bridge (1993), was which of the channel segments should be used to define channel sinuosity and which of the bars should be used to define channel splitting. This problem is highlighted when cross-bar channels are considered as it is debatable whether these low-stage features should be ignored when defining channel pattern although they may well evolve into major channels. Bridge (1993) noted that this problem has led to the development of a system that assigns orders to channels and bars in multiple channel rivers, then each order of channel is described by a particular flow pattern (Williams and Rust, 1968; Rust, 1978; Bristow, 1987; Thorne et al., 1993).

1.2.4 Channel Hierarchy and Braided Classification

Williams and Rust (1968) proposed a system of braided channel classification that incorporated three levels of channels and bars (figure 1.2a). Their first order channels formed the main anabranches of a braided system that flow around first order islands or bars. Second and third order channels in the Williams and Rust approach dissect these first order bars to form second and third order segments of bars. The distinction between the second and third order channels is unclear, the second and third order bars that they form are actually only dissected segments of first order bars, and these second and third order channels may well contain mid-channel bars of their own (Bridge, 1993).

In a study of channel migration and deposition on the Brahmaputra River, Bangladesh, Bristow (1987) defined a three level channel hierarchy system (figure 1.2b). The entire river system makes up the first order channel, with the channel margin being defined by the outermost river banks. First order channels may be comprised of several second order channels that form the main anabranches of the system, and were observed on the Brahmaputra to display a variety of channel patterns. These second order channels may themselves contain third order channels such as cross-bar channels. Bristow (1987) found that the bars around which the channels divide and rejoin, scale with the bankfull depth and width of each channel. Further work on the Brahmaputra River by Thorne et al. (1993) confirmed the existence of Bristow's channel hierarchy system. Thorne et al. (1993) went on to describe in more detail the nature of the hierarchy system and the way in which islands, bars and various bed features were scaled by the various channel orders.

Figure 1.2 Channel and bar ordering schemes of (a)Williams and Rust (1969), and (b) Bristow (1987). Numbers in circles refer to bars, other numbers refer to channels. Adapted from Bridge (1993).

Bridge (1993) put forward a 2 level channel and bar hierarchy system, in which the largest scale of bars or islands in a system and their adjacent channels are first order, but unlike Williams and Rust's scheme, all of the channels cutting across first order bars are second order. However, the dissected segments that the second order channels form are not themselves second order bars. Second order bars are defined as those which form in and at the termination of second order channels (figure 1.3).

Figure 1.3 Alternative channel and bar ordering scheme proposed by Bridge (1993).

1.2.5 Braided and Anastomosed Channel Patterns

Generally, the term braided when applied to river channel pattern is taken to mean the splitting of flow around an island or bar (Bridge, 1993). However, there is another mechanism of flow divergence widely recognised, that of anastomosing (Lane, 1957; Smith and Smith, 1980; Knighton and Nanson, 1993). An anastomosing river consists of multiple channels separated by islands which are usually excised from the continuous floodplain and which are large relative to the size of the channel (Knighton and Nanson, 1993). The principal characteristic of an anastomosing system is that each channel segment behaves significantly different from adjacent segments, having its own channel pattern with an independent sinuosity or braiding index (Bridge, 1993) and the distance between channel junctions is independent of channel width (Thorne et al., 1993).

However, in many multi-thread rivers the distinction between braided and anastomosed channel reaches is unclear (Bristow, 1987). For example, on the Brahmaputra River, Bristow (1987) found that some second order channels (defined using Bristow's classification scheme) contained many features characteristic of anastomosed channel reaches. Bridge (1993) proposed a solution to this problem by defining anastomosed reaches as those where the length of channel segments exceed the length of first order channels around individual first order bars (figure 1.4).


Lane (1957) described braided rivers as, 'having a number of alluvial channels with bars and islands between meeting and dividing again, and presenting from the air the intertwining effect of a braid'. The continual coming together and diverging of channels is inherent in the nature of braided systems (Bristow and Best, 1993). The flow dynamics and morphology of regions of channel convergence have been studied by many researchers including; Mosley (1976, 1982), Best (1988), Best and Roy (1991) and Roy and Bergeron (1990).

Figure 1.4 Various kinds of channel pattern as defined by Bridge (1993), adapted from 1994 Landsat-TM image of the Brahmaputra in the region of Bahadurabad

However, the connection between flow convergence and the subsequent downstream divergence of flow has often been disregarded (Bristow and Best, 1993), Ashworth (1996) being a recent exception. Ashmore (1991), Ashworth et al. (1992) and Bristow and Best (1993) have all commented that the region of flow divergence is the area that may well be fundamental to the development of braid bars.

1.3.1 Channel Confluences

The flow dynamics and morphology of channel confluences have received considerable attention recently in geomorphology, sedimentology and hydraulic literature (Mosley, 1976; Ashmore and Parker, 1983; Roy and Bergeron, 1990; Klaassen and Vermeer, 1988; Best, 1988). Where channels converge in a braided system, rapid changes occur in the velocity and sediment distributions that result in rapid shifting of the channel geometry (Richards, 1980).

Confluence channel morphology is widely accepted as being typically composed of three distinct elements that are controlled by the confluence angle and the ratio of discharges, these are; (1) avalanche faces at the mouth of each channel, (2) a deep confluence scour and (3) a bar formed in the separation zone at the downstream junction corner (Best, 1988). Best (1988) found that increases in the confluence angle or discharge ratio resulted in greater confluence scour, deflection of sediment around the increased zone of turbulence and flow separation at the confluence centre.

Models of confluence behaviour have traditionally assumed that the converging channels are of equal depth (Best, 1987) a condition that is rarely found in natural systems (Kennedy, 1984). Best and Roy (1991) suggested that a depth differential in two converging channels has a significant affect on the three-dimensional flow structures at the channel confluence and on the downstream mixing of the flows. Best and Roy (1991) further suggested that these depth-differential processes may explain the development of helical flow cells reported by Mosley (1976) and Ashmore (1982).

In flume experiments and field studies, Mosley (1976) observed that although the confluence angle and confluence scour at a particular junction varied in a consistent manner with relative discharge ratios and sediment loads, the overall morphology of a confluence can only be wholly understood when factors such as changes in flow orientation upstream of the channel confluence, bed material characteristics, structure and lithology are taken into account. Under certain conditions twin helical cells were observed downstream of the confluence, converging and downwelling at the channel centreline with outward near-bed flow towards the banks and inwards surface flow in both halves of the channel (figure 1.5).

Figure 1.5 Flow patterns in a model confluence; adapted from Mosley (1976)

An investigation into the affect of channel stage on confluence behaviour was conducted by Roy and Bergeron (1990) at a gravel-bed confluence on the Eaton North River, Québec, Canada. The results demonstrated that the observed flow structures in the channel confluence were sensitive to changes in flow stage. At low stages (less than 1/3 bankfull) the observed flow structures were controlled by bed features and in particular by the steep avalanche faces of the confluence scour. At higher stages the observed flow structures were controlled by the overall alignment of the confluence.

1.3.2 Flow Difluence

Anabranches of braided river systems shift rapidly across their braidbelt, constantly producing new channel junctions, sediment erosion through confluence scour, subsequent mid-channel bar formation and growth downstream (Ashworth, 1996). It is this process of bar formation and development that has received little quantitative attention in academic literature, although our understanding of the flow mechanisms and behaviour of channel confluences has been greatly improved in recent times. However, the subject of the downstream flow divergence in braided rivers has received little attention despite the fact that this region may well be fundamental to our understanding of bar genesis and evolution (Leopold and Wolman, 1957; Ashmore, 1991).

In general, areas of flow divergence within a fluvial system are associated with flow deceleration and sediment deposition. Once sediment deposition has been initiated, the sediment accumulation will probably promote further flow divergence, additional flow deceleration and sediment deposition and finally bar formation (Bristow and Best 1993). Flow being diverted around mid-channel sediment deposits will also impinge on the channel boundaries at an increased angle of attack. This results in bank erosion, widening of the braid belt and localised increases in the amount of sediment available for transport. All of these conditions are liable to lead to the development of a new braid bar (Carson 1984, Thorne et al., 1993). Bristow and Best (1993) commented that there was a clear need for a better understanding of the fluid dynamics and their effect on braid bar initiation in areas of flow bifurcation.

A recent attempt at rectifying this gap in our knowledge was made by Ashworth (1996), in a generic-scale flume study of mid-channel bar growth of a fixed junction scour. Ashworth's flume experiments differed from previous work by Leopold and Wolman (1957), and Ashmore (1982, 1991) in two respects; first, the channel junction was fixed and not allowed to develop naturally, Ashworth concluded that this probably had the effect of increasing the rate of morphological change; second, previous studies have involved 'before and after' type observation, whilst Ashworth continually monitored both flow structures and sediment transport. Ashworth (1996) presented a new-model for explaining the long-term development of the confluence-difluence unit (figure 1.6), the chain of channel events in Ashworth's model are:

(a) 'development of a confluence scour with flow convergence and maximum velocity in the channel centre;

(b) exceedence of the local transport capacity and initial stalling of coarse sediment in the channel thalweg downstream of the scour;

(c) bar growth through entrapment of all sizes of bed load;

(d) change from velocity maximum to minimum and flow convergence to divergence when the bar height is approximately 55 per cent of the thalweg depth;

(e) broadening of the bartop platform, a drop in local competence and bankward migration of the two distributaries whose cross-section and velocity remains approximately constant.'

Figure 1.6 (a) to (d) Ashworth's (1996) model of mid-channel bar growth downstream of a junction scour, arrows refer to surface flow direction.

1.3.3 Summary

Leopold and Wolman (1957) first recognised the importance of the confluence-difluence unit to braided river morphology. It has been shown by many authors since (Carson and Griffiths, 1987; Bridge, 1993; Goff and Ashmore, 1994; Ashworth, 1996) that a proper understanding of the flow and sedimentary processes involved in the confluence-difluence unit and the cycle of braid bar building is fundamental to our comprehension of braided river behaviour. Further studies are now required to combine the qualitative and quantitative data collected in flume experiments with macro-scale field studies of three-dimensional flow structures and sediment transport processes (Ashworth, 1996), in particular the response of all factors to the wide range of flow stage that characterise the flow regimes of most braided rivers.


The planform appearance of many braided rivers change dramatically with changes in flow stage. Bristow and Best (1993) noted that some authors have even suggested that these fluctuations in flow stage or discharge are a prerequisite for river braiding (e.g. Doeglas 1962, Miall, 1977), although this hypothesis has been discounted in many cases by the physical modelling of braided river planforms in steady discharge flume experiments (Ashmore 1982, 1991).

Miall's (1977) conclusion that fluctuations in discharge are a prerequisite for braiding stemmed from a study by Wright et al. (1974). Wright, Coleman and Erickson carried out a multivariate statistical analysis on thirty-four alluvial systems over the whole range of climatic zones and sizes. Four salient features of the discharge regime were compared: mean annual discharge, a coefficient of variability for mean monthly discharge, total discharge range and discharge flood peaks. An examination of their findings by Miall (1977) revealed a result that Wright et al. apparently overlooked. Their data showed that rivers dominated by braided channels have on average higher flood peaks, higher total discharge ranges and higher mean monthly discharge variability. Miall (1977) stated that these correlations were of critical importance to our understanding of the braiding process, and that the high variability in discharge ranges resulted in the processes that lead to bar initiation having a higher incidence of occurrence, notably stalling of the coarser sediment as discharge falls.

However, Ashmore (1982) reproduced the forms and processes of natural gravel braided rivers in constant discharge, constant slope flume experiments with equilibrium maintained using an adjustable sediment feed. This led him to conclude that the mechanisms of bar initiation proposed by Doeglas (1962) and Miall (1977) which rely on a variable or falling discharge are not uniquely responsible for braiding.

Further work has shown that bars may disappear at high flow stages, the implication being that some braided systems act as single channels at bankfull stage and only adopt the characteristic braided pattern on the falling stage (Bluck 1979, Smith 1974, Carson 1984, Gupta and Dutt 1989).

Carson (1984), in a study of a mechanistic approach to the classification of channel pattern in relation to a variety of river types in the Canterbury Plains, New Zealand, found that many large bars were in fact merely exposed tracts of the flat bed, dissected into lower order bed features, rather than a cluster of small genuine medial bars, and as a result at bankfull stages the channels revert to acting as single channels. Many of the other braided features of the rivers observed by Carson (1984) were in fact formed by low flow channels dissecting large point bars during flood events. The channels slowly revert back to their meandering pattern through gradual accretion of sediment in the ruts.

Gupta and Dutt (1989) found that the Auranga River, in the Bihar region of India behaved in large part as a braided river during low-season flows and as a single channel meandering river during high flow monsoon events. They observed that during monsoon flood events all of the braid bars were submerged and many were in fact missing as the water level reached the same height as the point bar. They suggested that many humid rivers with a strong seasonality in discharge and a large sand sized sediment load do fit the existing geomorphological models well.

Bristow and Best (1993) stated that these observations appear to be atypical and the majority of braided river systems, including humid Brahmaputra River, which have been observed to maintain their bars at both high and low flow stages (Coleman 1969, Smith 1970, Cant and Walker 1978, Church and Jones 1982, Bridge et al. 1986, Bristow 1987).

The consensus from the literature over the last 30 years seems to be that braided rivers react to changes in flow stage in the following manner: at high discharges or flow stages when the greatest amount of sediment is being transported, channels within the braided system are often scoured and mid-channel bars may be entirely eroded; during lower magnitude flow events maximum sediment deposition ensues, channel beds start to aggrade and any new high-stage bed-forms may be modified. It is during these low flow stages that new bars may be formed or existing bars may be enlarged as sediment is deposited; and as the flow stage falls further, some mid-channel bars may become emergent and be dissected by low stage channels, forming another braid level within the system (Bristow and Best, 1993).


A comparison of confluence scour in the braided sand-bed Brahmaputra river with that found in gravel-bed braided rivers carried out by Klaassen and Vermeer (1988) suggested similar relationships between scour hole depth and confluence angle. Klaassen and Vermeer looked at historical as well as their own field data. Historical data was provided by the Bangladesh Inland Water Transport Authority, who have been carrying out soundings of the Brahmaputra River bed since the late 1960s. They collected their own data during the monsoon floods of 1987. They found that a good relationship existed between scour depth and the angle of incidence of the anabranches, and that this relationship was very similar to that derived for gravel bed rivers by Ashmore and Parker (1983). In contrast to the Klaassen and Vermeer (1988) study, Simons and Simons (1987) concluded in a recent review of the differences between gravel and sand-bed rivers that 'gravel-bed reaches of a river system exhibit totally different morphological characteristics and in general, gravel bed river reaches will be less responsive to modest changes in discharge'.

Bristow and Best (1993) in their recent review of braided river issues and problems concluded that although there is an apparent differentiation necessary when calculating sediment transport parameters and considering sedimentary structures and detailed bar features, there would seem to be more gross channel morphological similarities between sand-bed and gravel-bed rivers than differences.

In geomorphological and sedimentological literature the distinction between gravel-bed and sand-bed rivers has been long established, despite the fact that it is unusual for a particular river to have a homogenous bed material. Though the apparent similarities between planform characteristics in gravel-bed rivers with that of sand-bed rivers suggest that they may share many important common processes, Bristow and Best (1993) found that little quantitative data exists to enable the comparison.


Bristow and Best (1993) commented that investigations into braided channel morphology and dynamics have covered the entire scale of braided river size, from micro-scale physical models of braided systems in the laboratory (Ashmore 1982) through small upland braided drainage channels (Smith, 1974; Ferguson, 1992), to meso-scale rivers with braid belts several kilometres in width (Ashworth et al. 1992, Warburton et al. 1993), to the largest macro-scale alluvial rivers such as the Brahmaputra (Coleman 1969, Bristow 1987, Thorne et al. 1993).

The problem of applying our understanding of key morphological processes from one of these channel sizes to braided channels of entirely different magnitudes has yet to be fully addressed (Bristow and Best, 1993). The apparent similarities of planform and cross-sectional characteristics require further investigation (Bristow and Best, 1993; Sapozhnikov and Foufoula-Georgiou, 1996). Similar results demonstrating close linkages between form and processes and spatial structure (Tarboton et al., 1988; La Barbera and Rosso, 1989; Nikora, 1991) in meandering channels across the whole range of channel scale have led to a better understanding of meander behaviour (Leopold, 1995).

1.6.1 Spatial Scaling in Braided Rivers

Sapozhnikov and Foufoula-Georgiou (1996) found quantitative evidence for this apparent similarity in spatial structure in braided rivers. Using a logarithmic correlation integral method they examined the spatial characteristics of three rivers (the Aichilik and Hulahula in Alaska and the Brahmaputra in Bangladesh) whose braid belt width varies from 0.5 - 15 km. The presence of similarities across scales in braided rivers implies that statistical properties of a braided river at one scale directly relate to the statistical properties of a braided river at another scale. Objects showing the same spatial scaling in all directions are called self-similar fractals. However, it is more usual for scaling properties to be different in different directions and such objects are called self-affine fractals. Self-similar fractals are characterised by one fractal dimension D, whereas, self-affine fractals are characterised by two fractal exponents: the local fractal dimension DL and the global fractal dimension DG (Mandelbrot, 1986). Sapozhnikov and Foufoula-Georgiou (1996) found that the three rivers exhibited anisotropic scaling or self-affinity in the downstream and cross-stream directions despite the large differences in braid belt width and types of bed material. They concluded that this result had important implications for the underlying processes responsible for the formation of the spatial structure of braided rivers.

1.6.2 Summary

The apparent scale invariance of the braided pattern would seem to have implications for the nature of the main processes involved in shaping braided morphology, bar genesis and evolution. As data becomes more available on larger rivers such as the Brahmaputra with advances in survey techniques, the issue of morphological processes and their relationship to channel scale in braided systems can start to be addressed.


It has long been established in academic literature that flow in alluvial rivers is strongly three-dimensional (Peters and Goldberg, 1989). Secondary currents were originally defined by Prandtl (1952) as currents which occur in the plane normal to the axis of the primary flow, they originate from interactions between the primary flow and gross channel features. Two types of secondary currents have long been recognised; skew induced and stress induced secondary currents.

Stress induced secondary currents develop in straight channels due to anisotropic distributions of boundary shear stress (Brundrett and Baines, 1964; Gessner and Jones, 1965; Perkins, 1970) and tend to be weaker than skew induced currents. Stronger secondary currents are caused by skewing of cross-stream vorticity into a longstream direction where there is a change in a channel parameter that can skew the primary flow, such as a channel bend or rapid change in cross-sectional shape (Perkins, 1970; Squire and Winter, 1951; Prandtl, 1952). Both types of secondary currents affect the primary velocity profile and the boundary shear stress distributions, therefore, they are of critical importance when considering flow in alluvial systems.

1.7.1 Secondary Currents in Meandering Channels

Over two decades of research has established that the pattern of primary isovels and pathways of bed material transport (both bed load and suspended load) in meandering rivers are strongly affected by skew-induced secondary currents. The influence of secondary currents on flow and sediment dynamics causes meander shifting through river bank erosion and bar sedimentation that leads to the planform evolution that is typical of meandering rivers (Friedkin, 1945; Thorne and Lewin, 1979; Lapointe and Carson, 1986).

Until the studies of Hey and Thorne (1975), and Bridge and Jarvis (1976) it was thought that secondary flow in meander bends consisted of a single cell of helical rotation (figure 1.7a), carrying surface water towards the outer bank and bed water towards the inner bank (Thomson, 1876; Hawthorne, 1951; Fox and Ball, 1968; Quick, 1974). Measurements taken by Hey and Thorne (1975) indicated the existence of a small cell of reverse rotation close to the outer bank and they proposed a new theory of secondary flow at meander bends incorporating this second outer bank cell. However, due to the lack of appropriate survey equipment the measurements made by Hey and Thorne (1975) and Bridge and Jarvis (1976) were relatively inaccurate and lacked detail. However, the development of the electromagnetic current meter that allowed measurements of perpendicular velocity components simultaneously, allowed more detailed measurements to be made.

Figure 1.7a Single cell theory of bed flow in a meandering river after Thomson (1876), Hawthorne (1951), Fox and Ball (1968) and Quick (1974).

Bathurst et al. (1977) reported on a study of secondary currents where measurements were made using an electromagnetic current meter at several cross sections perpendicular to the outer banks around three meander bends on the River Severn. At two of the three meander bends, Bathurst et al. found evidence for an outer bank cell extending up to one bank height distance away from the outer bank (figure 1.7b). The classical single helical cell was observed at all the meander bends. Bathurst el al. concluded that the presence of the outer bank cell depended on two factors, the shape of the outer bank; banks perpendicular to the water surface were more likely to result in an outer bank cell, and the strength of the component of flow towards the outer bank.

Figure 1.7b Current theory of bend flow with skew-induced and outer bank cells plus outwards flow at the inner bank after Hey and Thorne (1975), Bridge and Jarvis (1976), Bathurst et al. (1977, 1979).

Thorne and Hey (1979) described a study of secondary currents at an inflexion point between meander bends. This followed on from experiments carried out in flumes (Chacinski and Francis, 1952; Chacinski, 1954; Toebes and Sooky, 1967) to establish exactly how circulation from an upstream bend decayed and was replaced with the circulation associated with the subsequent downstream bend.

Toebes and Sooky (1967) found that the secondary circulation at the inflexion point consisted of two symmetrical cells, that converge at the surface and diverge at the bed, they suggested that the skew-induced cell for the downstream bend gradually displaces the relic cell from the upstream bend. However, the flume work of Chacinski and Francis found that the pattern of secondary flow at the inflexion point consisted of two stacked cells, the skew-induced cell from the downstream bend originating at the bed and displacing the skew-induced cell from the upstream bed. This apparent contrast in results may be explained by the width to depth ratios used by the two sets of workers. Chacinski and Francis used width to depth ratios that are much more likely to be found in natural systems.

Thorne and Hey (1979) used an electromagnetic current meter to investigate the secondary flow pattern at an inflexion point on the River Severn. They used three cross sections, starting just downstream of the first meander bend, continuing through the river inflexion point and ending just upstream of the second meander bend. They found that the pattern at the first cross section, just downstream of the first meander bend, was dominated by the skew-induced cell from the first meander bend, with some evidence of an outer bank cell still present. At the inflexion point they observed a stacked system of skew-induced cells, with the upper cell being a relic of the skew-induced cell from the upstream bend and the lower cell being the developing skew-induced cell from the downstream bend. At the downstream bend entrance they found that the secondary pattern was dominated by the associated downstream skew-induced cell. Thorne and Hey found that the system of secondary circulation proposed by Chacinski and Francis (1952) can exist in natural channels as well as in flume experiments. However, one important difference was that the bed topography of natural channels meant that an earlier and more complete replacement of remnant cells from the upstream bend took place.

Further work in the late 1970s and early 1980s demonstrated that the skew-induced secondary cell does not extend to the inner bank (Dietrich et al., 1979; Dietrich and Smith, 1983; Thorne et al., 1985, Markham and Thorne, 1992). Dietrich and Smith (1983) through the analysis of several sets of field and laboratory data found evidence for a substantial topographically induced alteration in the secondary flow pattern in the upstream portion of a meander bend (figure 1.7b). They found that shoaling over the point bar in the upstream part of a meander bend forced the high velocity filament towards the pool and outer bank. This is accomplished by the outward acting centrifugal force overcoming the inward acting pressure gradient force caused by a transverse water surface slope. The primary effect is outward flow over the whole flow depth over the upper point bar. This result had a major implication for the modelling techniques used to describe the flow patterns and evolution of topography in meander bends.

The assumptions used by Engelund (1974), Bridge (1977), Zimmerman and Kennedy (1978) and Odgaard (1981) in various mathematical models of flow in river bends, of outward flow at the surface and inward flow near the bed over the whole channel width, and how this secondary pattern varied with channel curvature and flow depth were not realistic. The observed pattern of outward flow over the point bar (Dietrich and Smith, 1983) in which the skew-induced secondary circulation was confined to 20-30% of the channel width where the flow was deepest, meant that particle force balance relationships, traditionally used to assume an equilibrium bed topography, could not be used to predict the net cross stream sediment transport, and that the convective accelerations induced by topographic effects should not be ignored (figure 1.7c).

Figure 1.7c Effect of the point bar crest on flow pattern and sediment sorting in a meander bend (adapted from Dietrich, 1982).

Smith and McLean (1984) went on to include topographically induced convective acceleration terms and local changes in channel curvature in a mathematical model of bend flow. They were able to reproduce the free surface geometry and boundary shear stress distribution measured in real alluvial systems. Smith and McLean's model was very similar to that of DeVriend and Geldof (1983) with a slight difference in the derivation of the governing equations. However, both models gave similar results in terms of shear stress distributions and free surface geometry when applied to similar conditions.

In the papers cited above, geomorphologists and river engineers have shown how the pattern of secondary flow affects the distribution of primary velocity and boundary shear stress. Where the flow plunges, primary isovels are compressed leading to a steeper velocity gradient near the bed and therefore, greater bed shear stresses. While areas of upwelling secondary flow show reduced primary velocity gradients close to the bed and therefore, reduced boundary stresses. Hence, by altering the bed shear stress distributions secondary currents play a strong role in determining the distribution of scour and fill around the channel perimeter. This effect can be used to explain the morphological evolution of meander bends in terms of erosion processes at the outer bank and the development of the point bar.

Close to the outer bank, the flow plunges where the skew-induced and outer bank cells converge at the surface. As a result primary isovels are packed near the bed and bank shear stresses are high in this region, promoting toe scour and undercutting of the bank (Thorne and Lewin, 1979). This often leads to mass instability and bank collapse that produces rapid bankline retreat (Thorne, 1978; 1982; Thorne and Osman, 1988). Failure blocks fall to the lower bank and bank toe, but their residence time there is short due to the high velocities and shear stresses imposed by the plunging flow. Once failed debris has been swept away, the flow again attacks the bed, toe and lower bank, again over-steepening it and generating further mass instability. In this way, the secondary and primary flows combine to produce aggressive and effective bank erosion capable of driving rapid and sustained bankline retreat.

Sediment transport in meander bends is also strongly influenced by the secondary flow pattern. A key process in meander bend development is the balance of forces acting on bed material particles, which is made up of the component of gravity acting and the inward acting drag force (Allen, 1970; Bridge, 1977; Dietrich and Smith, 1984; Parker and Andrews, 1985; Markham and Thorne, 1992). This balance of forces leads to an effective sorting mechanism because the component of gravity on a grain is equal to the cube of the particle diameter whilst the inward acting drag is proportional to the square of the particle diameter. The larger particles will tend to roll down the channel bed under gravity towards the thalweg, whereas the smaller particles will tend to be swept inward by the fluid drag resulting from the inward acting near bed secondary currents.

On the upper, point bar platform, the topographically induced outward flow transports bed material laterally outwards, towards the point bar crest. Sedimentation on the platform consists mostly of fines deposited due to decreasing flow discharge and velocity in the longstream direction. At the junction of the skew-induced cell and the zone of outward flow, near-bed currents converge and there is upwelling and a decrease in the near bed shear forces, promoting sediment deposition in this region. This accumulation of sediment leads to a building a of a sharp ridge separating the upper, gently sloping point bar platform from the lower, steeper point bar face.

In meandering rivers the point bar crest follows the zone of bed load convergence, which is skewed across the channel from the outer bank at the bend entrance, to the inner bank at the bend exit (Dietrich et al., 1982). This topographic feature itself induces further strong circulations that play important roles in sediment sorting by size fraction and determine sediment pathways through the bend (Dietrich et al., 1979; 1984; Dietrich and Whiting, 1989, Markham and Thorne, 1992)

There is now overwhelming evidence that secondary currents significantly affect channel morphology in single-thread meandering rivers and that recognition of the pattern of secondary currents (both helical cells and lateral flows) are crucial to understanding, explaining and modelling flow and sediment processes in channel bendways.

1.7.2 Secondary Currents and Channel Changes in Braided Channels

There have been few field and laboratory investigations of flow structures in braided rivers and as a consequence their link to braided morphology is little understood. Unlike meandering single-thread rivers in which observations of key flow processes can be undertaken over a range of flow stages, in braided rivers most channel changes are associated with changes in bed morphology with occur at high discharges when observation is very difficult (Smith, 1974; Rust, 1978; Ashmore, 1982). As a result of the difficulty of studying flow structures in the braided environment, research to date has relied heavily on laboratory investigations. In addition, any mention of secondary currents in braided systems has been restricted to areas of channel confluence, and the effect of secondary currents in bifurcations and around braid bars has been largely neglected.

Mosley (1976) reported the presence of helical flow at channel confluences in a general flume experiment of confluence behaviour. Using dye injection to visualise flow patterns, Mosely observed that the pattern of secondary flow at channel confluences consisted of a pair of helical cells, converging in the channel centre over the point of maximum scour and diverging at the bed. Mosley concluded that the observed helical flow pattern resulted in steepening of the scour walls beyond their natural angle of repose, giving rise to their characteristic avalanche faces. Mosley further stated that the helical flow structure resulted in most of the sediment transport in the confluence occurring at the channel fringes away from the area of maximum scour. There was also evidence found for smaller cells of reverse rotation further downstream from the channel confluence, resulting in elevated portions of the bed flanking the channel centreline. Mosely observed that this secondary flow pattern resulted in high rates of sediment transport restricted to the zones between opposing cells (figure 1.5).

In another laboratory investigation, Ashmore (1982) successfully reproduced gravel-bed braided stream morphology in a small scale constant discharge and slope flume experiment. Again using dye injection as a visualisation tool, Ashmore reported similar patterns of secondary flow associated with channel confluences and scour to those observed by Mosley (1976). Ashmore found that maximum scour at the channel confluence occurred in the zone of attachment of the helical cells and deposition occurred in the zone of separation of the cells.

Best and Roy (1991) proposed that these secondary flow patterns are the result of horizontal separation vortices formed in the lee of the avalanche faces at the entrance to the confluence, particularly when the converging channels are of unequal depth. Another probable cause of the helical circulation is the same mechanism that causes helical flow in meander bends, i.e. the relationship between the outwardly directed centrifugal force and the inwardly directed pressure gradient force caused by super-elevation. In the case of channel confluences the centrifugal force would be acting towards the confluence centre and the pressure gradient force would be acting towards the confluence fringes. Mosley (1976) and Ashmore (1982) both observed a degree of superelevation in the centre of their model confluences. Ashmore et al. (1992) proposed that the flow separation at the confluence entrance probably reinforced, rather than replaced, the circulation due to channel curvature.

Ashmore et al. (1992) carried out a field investigation of secondary flow patterns in river confluences on the gravelly Sunwapta River, Alberta. Ashmore et al. confirmed for the first time in a real river the existence of the secondary flow patterns observed in laboratory experiments by Mosley (1976) and Ashmore (1982). Ashmore et al. made measurements in two recently formed Y shaped anabranch confluences at relatively high discharge levels. Ashmore et al. found that the pattern of secondary flow observed in laboratory experiments existed in both of the surveyed cross-sections, although the pattern was much stronger in one of the confluences. In both confluences the larger of the two helical cells tended to dominant the other in the downstream direction.

Unlike in the laboratory where dye injection can be used to visualise flow structures, field measurements require post survey correction of measured velocities. Ashmore et al. (1992) conceded that the methods employed to correct velocities in meandering rivers were inappropriate for braided rivers and as a result the flow patterns reported may be susceptible to small errors.

The issue of secondary flows around braid bars has received little attention, with few field investigations undertaken. Ashworth et al. (1992) in a descriptive model for bedload transport and sorting around a braid bar proposed a secondary flow structure similar to that found in meandering channels. They suggested that the actual pattern of secondary flow in the two anabranch channels around a braid bar would mirror the pattern in two 'back-to-back' meanders (figure 1.8). This system of flow structure in anabranches around braid bars taken together with the observed patterns of secondary flow in channel confluences (Mosley, 1976; Ashmore, 1982; Ashmore et al., 1992) would lead to the following impact of secondary flow on the morphology of braid bars;

(a) Downstream of a channel confluence a double helical flow is setup with flow convergence at the surface and divergence at the bed. Bedload transport is concentrated in a narrow zone close to the channel centre,

(b) as the channel widens bar aggradation would be initiated, the double helix flow pattern would drive the finer sediment towards the outer banks and down the anabranch channels, due to the relative weakness of the secondary flow compared to the primary flow, the coarser sediment is not affected in the same way and is therefore often deposited on the barhead,

(c) a cross over of the secondary flow pattern would then occur in the anabranch channels, from the double helix pattern at the flow bifurcation to single helical cells rotating in the opposite direction similar to those reported in meandering channels.

(d) this pattern would now resemble two back-to-back meanders, with eroding outer banks, deep thalwegs and a pronounced point bar. Fine sediment would then be swept inwards by the single helical flow cell, collecting on the bar tail.

In braided rivers, the width to depth ratio tends to be higher than for single thread meandering river, and as such the secondary flow patterns fundamental to this model for bedload transport and sorting around a braid bar, may only develop in above average discharges (Ashworth et al., 1992).

Figure 1.8 Concept of flow in a bifurcated channel as mirror image meanders (Ashworth et al., 1992)




Braided rivers are highly dynamic systems characterised by high rates of erosion, substantial sediment transport and deposition. Despite the abundance of braided river systems and their importance to a whole range of disciplines including geomorphologists, sedimentologists, geologists and environment scientists, they have not been as extensively studied as single-channel rivers and river networks. This is probably due in part to the difficulty of undertaking fieldwork in this rapidly changing environment.

The studies to date that have been carried out to establish models and frameworks for understanding braided river behaviour have been mostly qualitative in nature. The lack of quantitative studies of braided rivers has impeded the development of our understanding of this complex environment.

There still remains a great deal of uncertainty over the actual classification of the braided system, with many conflicting definitions being used (Leopold and Wolman, 1957; Rust, 1978; Knighton and Nanson, 1993; Robertson-Rintoul and Richards, 1993; Bridge, 1993). As a result there is no clear distinction between mid-channel bars and islands, the relationship between channel stage and the braided planform has not been fully investigated and the precise distinction between braided rivers and anastomosed channels is unclear.

Another important area of study yet to be fully explored by quantitative investigations are the mechanisms and processes of channel confluences and difluences. It has been demonstrated by many authors that a proper understanding of the flow and sedimentary processes at channel junctions is fundamental to our comprehension of braided river morphology (Leopold and Wolman, 1957; Carson and Griffiths, 1987; Bridge, 1993; Goff and Ashmore, 1994; Ashworth, 1996). It is now essential that the qualitative and quantitative data that do exist from flume studies and small scale field studies be combined with large scale field investigations of three-dimensional flow structures and sediment processes, in particular the response of these to the wide range of flow stages that characterise most braided rivers.

Within the literature over the last 30 years there seems to be a good qualitative agreement of how braided rivers react to changes in flow stage. At high flows when most sediment is being transported, channels are scoured and mid-channel bars are often entirely eroded, whereas, during low stage events when maximum sediment deposition occurs, channel beds aggrade and new bars are formed or existing bars may be enlarged. However, more quantitative data is needed on the processes involved during this cycle of flow stages.

The apparent similarities between large scale and small scale braided rivers in terms of spatial characteristics and gravel and sand-bed braided rivers in terms of morphology has important implications for the main processes involved in shaping braided river morphology, bar formation and evolution. Although attempts have been made at a quantitative analysis of the scale invariance of planform characteristics (Sapozhnikov and Foufoula-Georgiou, 1996) and the influence of bed-material type on braiding (Klaassen and Vermeer, 1988), there still remains little quantitative data to enable examination of similarities.

Despite that fact that is has long been established that flow in alluvial channels is strongly three-dimensional and that there is now overwhelming evidence that secondary currents significantly affect channel morphology in single-thread rivers , there have been few field and laboratory investigations of flow structures in braided channels. This is due in part to the fact that most channel changes in braided systems are associated with changes in bed morphology which occur at high discharges, when observation can be difficult (Smith, 1974; Rust, 1978; Ashmore, 1982). As a result most studies of flow structures have been limited to laboratory investigations. The field studies that do exist have tended to concentrate on areas of channel confluence. The effect of three-dimensional flow structures at channel bifurcations and around braid bars has been largely neglected.